In this episode of Deal Volume, McKinsey’s podcast on private markets, McKinsey partner and host Brian Vickery speaks with senior partner Ben Ellencweig. Ben leads alliances and partnerships for QuantumBlack, AI by McKinsey, and spends much of his time working with private equity (PE) firms and their portfolio companies. In this discussion, they get beyond the hype surrounding generative AI (gen AI) and discuss its implications for PE stakeholders. An edited version of their conversation follows.
How gen AI can take PE firms to the next level
Brian Vickery: Let’s start by clarifying a few definitions. Companies have been using machine learning [ML] and analytical AI for several years now. How does gen AI differ from those technologies?
Ben Ellencweig: Analytical AI and ML algorithms are used to complete analytical tasks—classifying huge amounts of data, predicting the cluster, and evaluating it—faster and better than humans, which makes them great for tasks such as customer segmentation, sentiment analysis, and sales forecasting. Gen AI can take creating content to the next level—for example, new lyrics and music for a Beatles-style song or software code—as if a human was involved. The content can be images, videos, or text. People think about content generation in the context of marketing, but it could be creating a technician manual, assisting in drug discovery, or enhancing efficiencies for a portfolio company.
Brian Vickery: There’s an enormous amount of hype about gen AI. What do you characterize as hype versus potential, and where are people actually using gen AI today?
Ben Ellencweig: We see four big use case archetypes today. The first is code generation: there are code and developer productivity tools and copilots to help us write better code, run quality assurance on it, and make sure we’re generating enough synthetic data as we create software. The second is content generation, such as for marketing materials—including the holy grail of hyperpersonalized communication—and technical manuals. The third is human engagement, using bots or agents to create a new experience in customer service, sales enablement, or servicing employees in finance and HR functions. And the fourth is the virtual knowledge worker that can summarize and extract insights from large amounts of information, including unstructured data sources. These are still the early days for some uses cases. Others are more advanced, and we’re seeing innovation day by day.
Brian Vickery: How are most companies approaching these archetypes? Are they initially trying out one? Is anyone adopting all four?
Ben Ellencweig: Yesterday, I participated in a forum with about 65 PE operating partners. I asked how many have portfolio companies that are adopting gen AI, and about 60 percent put their hands up. Then I asked this subset if their companies are not only experimenting but also in production at scale, and only three hands, or about 5 percent, stayed up. I would say this is typical of what we see.
Interestingly, before gen AI, most companies failed with advanced analytics or AI transformations because it’s hard. It requires a lot of change management to win hearts and minds. You have to measure impact. Making investments calls for patience while waiting for the ROI. Data governance is a major concern, and the list goes on. Gen AI adds even more layers of complexity. If 2023 was the year of experimentation with pilots everywhere, 2024 and especially 2025 will be years of scaling and actually moving impact to the bottom line.
The PE industry can learn from other industries on how to leverage gen AI
Brian Vickery: What questions do PE firms have about gen AI, and how are you advising them?
Ben Ellencweig: Three questions come up a lot. First, PE firms want to know what others in their segment are doing—for example, to create efficiencies—and how they compare. Earlier today, I met with a portfolio company in the drug discovery space, and at one point, they asked whether any other companies in their sector were doing what they are doing. The answer was maybe, but that’s not the right question. Instead, they should ask how a German automotive OEM or an aerospace developer is thinking about using gen AI and digital twin–simulation tools in R&D and then apply those lessons to drug discovery. It’s important to go beyond the cutting edge in your industry and understand which innovations are transferable across industries.
Then they want to know how to get started. As I said, this is complex and involves not only technology but also a lot of psychology. The portfolio management team has to be on board, and there are decisions about where the budget comes from. Beyond that, we always talk about “two by two”: first, find two small use cases and just get going. It could be something straightforward such as an off-the-shelf bot solution using gen AI in the call center. But start as soon as possible, and let people dabble with it. Your sellers, buyers, and customers are smart; they will see what works and adopt it pretty quickly. This creates the right momentum in the short term.
In parallel, think about two strategic workflows for the company—the essence of who you are and what you do. For example, an industrials company was considering developing an R&D copilot because they were charging a premium for their product. But the CEO said, “No, we’re charging a premium because we’ve got great service. When Bill or Joe show up on the factory floor, they know the layout, the throughputs of the machine, and the history of repairs for the past 30 years. That is our value.” So instead, they developed a service copilot that helps technicians understand the history, find the right repairs, and order the right parts based on pictures taken with a mobile phone. It might take a year or more to reinvent a strategic workflow with gen AI, but that’s the two-pronged approach we recommend.
Finally, people ask how to measure progress so they can assess whether they are doing well or falling behind. It’s important to set a baseline and understand the problem you’re trying to solve so you can see if an improvement trickles to the bottom line.
A gracious approach to funding gen AI projects can yield meaningful returns
Brian Vickery: How have you noticed the perspectives between portfolio operations partners, portfolio company CEOs, and deal teams differ on gen AI, especially the budgeting for it?
Ben Ellencweig: Although everybody understands gen AI’s importance, aligning those three groups is quite challenging. Many times, the company is excited, but they don’t get relief from the owners to fund it. Or the PE owners and the operating partner are excited, but the company and investment partners are not.
Keep in mind it really isn’t that costly. By my estimates, roughly 1.0 to 1.5 percent of companies’ current IT budget would nicely cover the costs to deploy gen AI. That’s not including cloud and personnel, but the day-to-day opex [operational expenditure] is not massively more than what companies are currently spending on tech. I encourage owners or portfolio company executives to graciously provide funding because, although it’s not a big amount, it can make a huge difference because it helps avoid distorted decision making and can mean a lot for a company’s long-term health. It has a much faster ROI than any previous transformational IT project.
How gen AI is influencing dealmaking and due diligence
Brian Vickery: When you are working on diligence on behalf of PE firms, what’s the right way to take gen AI into account?
Ben Ellencweig: It reminds me of the early days of cyber. Assessing cyber during due diligence started with a single question: Is there any cyberthreat? Then the checklist became a page in the booklet that comes out of diligence. Now it’s a whole chapter.
Today, many people focus on productivity gains. But what about the revenue enhancement opportunities with gen AI? Gen AI could enable self-serve tools, new products, or enhancements to existing products that increase their value. Another angle is much more strategic. Will gen AI cause a tectonic shift in a company’s industry or in, say, its pricing? IT services and law firms, among others, charge by the hour, but that model is challenged by gen AI. If BrianCo cannot show productivity gains, the customer might prefer to hire BenCo. Yet if BenCo is still charging by the hour, the customer is losing money despite the efficiency.
It requires rethinking the three lenses of productivity, revenue enhancement, and strategic shifts, which affect not only business models but also pricing. Our diligence on gen AI today is roughly a page, but before long, it will be a chapter.
Practical considerations for getting started with gen AI in PE
Brian Vickery: Many PE companies feel their data is not in good enough shape for gen AI. What do you say to those folks?
Ben Ellencweig: Show me a company that doesn’t have massively messy data sets. This is a very common starting point. But when it comes to the data, the beauty is that you can use gen AI to clean up your data to make it usable for other gen AI applications to extract value.
To get started, remember technology is only 20 percent of the work. I envision three prongs. First, get organized and put in place all the enablers. That includes getting the data in order, setting up data governance, and identifying technology partners and talent needs. It also includes considerations about structure. You can set up a center of excellence at the fund level for all the portfolio companies, hire a third party, or tell the portfolio companies to do it for themselves. You also need a champion. I talked to a guy yesterday who said he’s spending 80 percent of his time on his gen AI “sidekick” even though he’s an operating partner.
Next, think about psychology and change management. Educate your colleagues. Run a “gen AI day” for the CEOs, CIOs [chief information officers], and CTOs [chief technology officers] of your portfolio companies so they can share what they’re doing. Send them tutorials, and bring in outside speakers. This space is moving so fast that it’s hard for a day-to-day manager to keep up. And that’s where the fund can help. Also, educate your own colleagues at the fund level. The operating partners can make sure everyone stays in touch with what’s happening with technology. Third, there are portfolio-level applications, which require a road map. Having a plan is important because without it you can’t move forward.
Brian Vickery: We’re seeing a lot more reporting recently on the risks of gen AI. How do you reconcile where the world is going, and how we do get there responsibly?
Ben Ellencweig: This is new, and we’re still exploring the risks and how to mitigate them. Regulated industries such as financial services and healthcare have their own distinct compliance-related risks. There are safety risks in drug discovery, automotive, and aerospace, for example. We’re all working to mitigate risks of hallucinations—which is when the model thinks something is real but it absolutely is not—and biases. But we don’t talk as much as we should about the carbon footprint of those massive data centers or even how the United States will meet demand for electricity and AI chips for electric vehicles and gen AI.
I heard a great analogy: gen AI is like an intern. You give them a task, they go off, come back all excited, and create something that looks really good. It’s structured, thoughtful, and well organized, but it could be totally wrong. We need to look at the outputs of gen AI with a critical eye and apply our human judgment to evaluate whether we trust them. In many cases, the outputs are spot on and can advance our thinking. As long as we keep humans in the loop, we can do a lot of good with gen AI.
